Update app.py

app.py

@@ -3,9 +3,11 @@ from huggingface_hub import InferenceClient
 
 client = InferenceClient("Futuresony/future_ai_12_10_2024.gguf")
 
-def format_alpaca_prompt(user_input, system_prompt):
+def format_alpaca_prompt(user_input, system_prompt, history):
     """Formats input in Alpaca/LLaMA style"""
+    history_str = "\n".join([f"### Instruction:\n{h[0]}\n### Response:\n{h[1]}" for h in history])
     prompt = f"""{system_prompt}
+{history_str}
 
 ### Instruction:
 {user_input}
@@ -15,7 +17,7 @@ def format_alpaca_prompt(user_input, system_prompt):
     return prompt
 
 def respond(message, history, system_message, max_tokens, temperature, top_p):
-    formatted_prompt = format_alpaca_prompt(message, system_message)
+    formatted_prompt = format_alpaca_prompt(message, system_message, history)
 
     response = client.text_generation(
         formatted_prompt,
@@ -27,6 +29,8 @@ def respond(message, history, system_message, max_tokens, temperature, top_p):
     # ✅ Extract only the response
     cleaned_response = response.split("### Response:")[-1].strip()
     
+    history.append((message, cleaned_response))  # ✅ Update history with the new message and response
+    
     yield cleaned_response  # ✅ Output only the answer
 
 demo = gr.ChatInterface(
@@ -39,6 +43,5 @@ demo = gr.ChatInterface(
     ],
 )
 
-if 
-__name__ == "__main__":
+if __name__ == "__main__":
     demo.launch()